Pith. sign in

REVIEW 2 major objections 5 minor 69 references

Enforcing world-frame agreement on position, shape, and appearance of matched 3D Gaussians removes the geometric artifacts that pure 2D photometric losses leave in sparse-view reconstruction.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-14 09:12 UTC pith:TSSHBSTO

load-bearing objection Solid empirical regularizer for sparse-view 3DGS: multi-attribute world-frame consistency on MASt3R+DINOv3 matches delivers real wide-baseline gains, with the usual CV caveats. the 2 major comments →

arxiv 2607.10792 v1 pith:TSSHBSTO submitted 2026-07-12 cs.CV

MAC-Splat: Multi-Attribute Consistency for High-Fidelity Sparse-View Reconstruction

Jinqian Yang , Yichen Wu , Wanhua Li , Haokun Lin , Renzhen Wang , Xiangchu Feng , Xixi Jia This is my paper
classification cs.CV
keywords 3D Gaussian SplattingSparse-View ReconstructionGeometric ConsistencySemantic GuidanceMulti-Attribute ConsistencyGeneralizable Neural RenderingMASt3RDINOv3
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Sparse-view 3D reconstruction is ill-posed: when only a few images with limited overlap are available, photometric rendering losses cannot resolve depth and correspondence ambiguities, so generalizable 3D Gaussian methods produce floaters, distorted shapes, and ghosting. MAC-Splat claims the missing signal is direct 3D consistency. It first builds high-confidence 2D matches with a geometric backbone enriched by frozen semantic features, then lifts those matches into world space and forces the associated Gaussians to agree on position, covariance shape, opacity, and color. On ScanNet++ this yields large gains over strong baselines—more than 4.5 dB average PSNR over Splatt3R—and the advantage grows as viewpoint separation increases. A sympathetic reader cares because the same sparse, unposed inputs arise in AR/VR, robotics, and casual capture; a training objective that actually ties corresponding 3D primitives together is a concrete way to make feed-forward Gaussian reconstruction reliable under those conditions.

Core claim

A direct multi-attribute 3D consistency objective, when anchored by high-quality semantically informed correspondences, is effective for the ill-posed sparse-view problem. By jointly regularizing matched Gaussians’ world-space position, log-eigenvalue shape, and appearance, MAC-Splat produces geometrically coherent reconstructions that maintain quality as camera pose gaps widen, improving average PSNR by more than 4.5 dB over Splatt3R while cutting LPIPS.

What carries the argument

The Multi-Attribute Consistency (MAC) loss: after reciprocal nearest-neighbor matches are filtered by confidence, each match selects a pair of Gaussians that are transformed into a common world frame; a confidence-weighted sum of Huber position error, Huber log-eigenvalue shape error, and L1 appearance error forces those attributes to agree.

Load-bearing premise

The method assumes that the sparse 2D matches from residual-fused geometric and semantic features correctly identify corresponding 3D Gaussians whose attributes should be forced to agree; if those anchors systematically link the wrong primitives, the loss regularizes the wrong pairs.

What would settle it

On a held-out ScanNet++ wide-baseline split, replace the predicted correspondences with ground-truth depth-reprojected matches (or deliberately inject controlled mismatch rates) and measure whether the reported PSNR/LPIPS gains over the photometric-only ablation disappear once the anchors are no longer reliable.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. MAC-Splat addresses sparse-view generalizable 3D Gaussian Splatting by adding direct 3D multi-attribute consistency supervision on top of a MASt3R backbone. Semantically informed 2D correspondences are obtained by residual fusion of frozen DINOv3 features into MASt3R descriptors, followed by reciprocal nearest-neighbor matching and confidence gating. These matches anchor the Multi-Attribute Consistency (MAC) loss, which compares world-frame position (Huber on centroids), shape (Huber on log-eigenvalues of covariances), and appearance (opacity and SH coefficients) of the associated Gaussians. Training uses photometric losses plus a fixed-weight MAC term; ground-truth poses are used only during training, while MASt3R poses are used at test time. On ScanNet++ the method reports large gains over Splatt3R and other generalizable 3DGS baselines across four overlap regimes, with especially strong retention of PSNR/LPIPS under wide baselines, plus supporting masked-metric and zero-shot DTU results and ablations that isolate MAC versus DINOv3.

Significance. If the reported gains hold under independent reimplementation, the paper supplies a concrete and practically useful recipe for sparse-view generalizable 3DGS: high-quality 2D anchors plus an explicit multi-attribute 3D regularizer that is rotation- and scale-aware and has an anti-degeneracy gradient. The log-eigenvalue shape term and the masked-metric evaluation on jointly observable regions are particularly valuable contributions; they move beyond pure photometric supervision and give a clearer reading of geometric fidelity under low overlap. The work is incremental relative to MASt3R/Splatt3R-style pipelines, but the combination of semantic residual fusion, multi-attribute world-frame consistency, and systematic wide-baseline evaluation is a solid advance for an important practical regime.

major comments (2)
  1. Sections 3.1–3.2 and Eqs. (1)–(7): the central claim that MAC improves geometry rests on the assumption that residual-fused reciprocal NN matches (after confidence gating) correctly pair Gaussians whose world-frame attributes should agree. The paper already shows that MAC without DINOv3 recovers most of the wide-baseline PSNR (Table 4: 20.23 vs. 21.34 dB on Very Wide) and that masked gains remain large (Table 3: +6 dB vs. Splatt3R on Very Wide). Still missing is a direct diagnostic of match quality under the same (ϕ,ψ) splits—e.g., fraction of retained matches that are geometrically consistent with GT depth/pose, or a controlled experiment that injects controlled match noise. Without that, residual risk remains that MAC is sometimes regularizing mis-associated primitives, especially in textureless or wide-baseline regions.
  2. Table 1 and Section 4.1 inference protocol: baselines are evaluated under their authors’ official protocols, which differ in pose requirements (SfM poses for PixelSplat/MVSplat vs. pose-free for Splatt3R/NoPoSplat/MAC-Splat). The large drop of PixelSplat/MVSplat on Very Wide is therefore not fully isolated from pose-estimation differences. A controlled re-evaluation of at least one posed baseline under MASt3R-estimated poses (or of MAC-Splat under GT poses for the context views) would make the robustness claim cleaner and more load-bearing.
minor comments (5)
  1. Eq. (5) and surrounding text: clarify that only the three eigenvalues are compared (order-sorted) and that the formulation is intentionally invariant to orientation of the principal axes; a short remark on whether orientation consistency is ever desirable would help.
  2. Implementation details (Section 4.1): report the number of retained matches per pair after τ_conf gating, and the sensitivity of final metrics to τ_conf and to λ_MAC (currently fixed at 0.25). Even a short appendix table would strengthen reproducibility.
  3. Figure 1 and Figure 6 captions: the qualitative claims about floaters and thin structures are persuasive, but adding a small inset or arrow highlighting the specific failure modes of the ablated variants would make the visual argument tighter.
  4. Related Work: a few recent concurrent sparse-view / pose-free 3DGS works (e.g., FreeSplat, InstantSplat, AnySplat) are cited but only lightly contrasted; a one-sentence positioning relative to their consistency mechanisms would improve completeness.
  5. Typographical / formatting: occasional missing spaces after commas in the abstract and introduction (“covariancematrices”, “leadstostable”); also “DINOv3” is introduced without a citation until the methodology section—move the reference earlier.

Circularity Check

0 steps flagged

No circularity: MAC loss is an independent regularizer evaluated on held-out novel-view metrics; nothing reduces by construction to its inputs.

full rationale

The paper defines 2D correspondences via residual-fused MASt3R+DINOv3 descriptors (Eq. 1, confidence gating), lifts them to world-frame Gaussians using known/estimated poses, and applies an independently specified multi-attribute loss (positional Huber on centroids, log-eigenvalue shape, L1 appearance; Eqs. 2–7) whose scalar weights are fixed once and never fitted to the reported metrics. Training combines this with standard photometric+LPIPS losses; evaluation uses ordinary PSNR/SSIM/LPIPS on held-out ScanNet++ difficulty splits and zero-shot DTU. Ablations (Table 4) and masked metrics (Table 3) further isolate the contribution without any self-referential prediction or uniqueness claim. No self-definitional equations, fitted-then-predicted quantities, load-bearing self-citations, imported uniqueness theorems, smuggled ansatzes, or renamed known results appear. The derivation chain is therefore self-contained and non-circular.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 3 invented entities

The central empirical claim rests on a small set of hand-chosen loss weights, the assumption that MASt3R+DINOv3 matches are sufficiently accurate anchors, and the modeling choice that bilinear sampling of dense Gaussian parameter maps yields meaningful corresponding primitives. No new physical entities are postulated; the invented items are purely algorithmic constructs whose value is measured by the reported metrics.

free parameters (4)
  • λ_MAC = 0.25
    Overall weight of the multi-attribute consistency term relative to photometric loss; fixed at 0.25 for all experiments.
  • λ_p, λ_s, λ_a = 1.0 / 0.02 / 0.1
    Relative weights inside MAC for position, shape, and appearance; set to 1.0 / 0.02 / 0.1 without further search reported.
  • τ_conf
    Confidence product threshold that gates which matches enter the MAC loss; paper asserts robustness but does not publish the numeric value used.
  • AdamW learning rate = 1e-5
    Optimizer step size fixed at 1e-5 for the Gaussian head and fusion MLP.
axioms (4)
  • domain assumption Reciprocal nearest-neighbor matches on residual-fused MASt3R+DINOv3 descriptors, after confidence gating, identify pairs of Gaussians that ought to share world-frame attributes.
    Load-bearing premise of Sections 3.1–3.2; if systematically false the MAC gradients become noise.
  • domain assumption Ground-truth camera-to-world transforms available at training time correctly place both Gaussians of a match into a common Euclidean frame.
    Stated in Section 3.2; test-time uses MASt3R poses instead.
  • ad hoc to paper Comparing logarithms of covariance eigenvalues yields a rotation- and uniform-scale-invariant shape metric that also repels geometric degeneracy.
    Equations (5) and (8); the anti-degeneracy claim is derived but not independently validated outside the training runs.
  • domain assumption Standard photometric ℓ2 + LPIPS losses on rendered novel views remain a valid primary objective when augmented by MAC.
    Equation (9); conventional in the 3DGS literature the paper builds upon.
invented entities (3)
  • Multi-Attribute Consistency (MAC) loss no independent evidence
    purpose: Jointly regularize position, log-eigenvalue shape, and appearance of matched Gaussians in world coordinates.
    Core algorithmic contribution defined in Section 3.2; value is measured only by the paper’s own ablations and tables.
  • Residual Semantic Fusion module no independent evidence
    purpose: Inject frozen DINOv3 features into MASt3R decoder descriptors via a small residual MLP.
    Described in Section 3.1; ablation shows it improves LPIPS and thin-structure fidelity, but the module itself has no external validation.
  • Semantically-Guided Geometric Consistency paradigm no independent evidence
    purpose: Framing that semantic features should resolve geometric ambiguities under sparse views.
    Stated as a contribution in the introduction; essentially a name for the overall pipeline rather than an independently testable object.

pith-pipeline@v1.1.0-grok45 · 20178 in / 3112 out tokens · 54012 ms · 2026-07-14T09:12:54.255622+00:00 · methodology

0 comments
read the original abstract

Reconstructing high-fidelity 3D scenes from sparse-views remains a central problem in generalizable neural rendering. Existing generalizable 3D Gaussian Splatting (3DGS) methods often exhibit geometric artifacts in sparse-view settings, since supervision based solely on 2D photometric losses cannot resolve depth and correspondence ambiguities. To address this issue, we propose MAC-Splat, a training framework built around direct 3D consistency supervision. MAC-Splat builds on the MASt3R geometric backbone and a frozen DINOv3 encoder to obtain semantically informed 2D correspondences, which serve as geometric anchors for 3D supervision. Using these anchors, we define the Multi-Attribute Consistency (MAC) loss. This objective jointly regularizes the 3D attributes of matched Gaussians, including their position, shape, and appearance, by enforcing agreement in a common world coordinate frame. The formulation is robust to outliers and respects the geometry of covariance matrices, which leads to stable training under sparse-view conditions. Experiments on ScanNet++ show that MAC-Splat outperforms strong baselines, with particularly large gains under different overlap regimes. In particular, it improves average PSNR over Splatt3R by more than 4.5 dB, reduces LPIPS, and maintains performance as the camera pose gap increases. These results indicate that a direct, multi-attribute 3D consistency objective, when combined with high-quality correspondences, is effective for addressing the ill-posed sparse-view reconstruction problem.

Figures

Figures reproduced from arXiv: 2607.10792 by Haokun Lin, Jinqian Yang, Renzhen Wang, Wanhua Li, Xiangchu Feng, Xixi Jia, Yichen Wu.

Figure 1
Figure 1. Figure 1: Matched-pixel constraints for 3D Gaussians. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Architectural overview of MAC-Splat. The framework augments a MASt3R backbone with features from a DINOv3 encoder through a residual fusion module to produce semantically enriched descriptors. A multi-stage filtering procedure yields sparse, high-confidence 2D correspondences. These matches anchor the Multi￾Attribute Consistency (MAC) loss, which compares the world-space position, shape, and appearance of … view at source ↗
Figure 3
Figure 3. Figure 3: From matched pixels to MAC loss. Each matched pixel pair selects two Gaussians in world space, and the MAC loss enforces consistency in their position, shape, and appearance. we compare the logarithms of their eigenvalues. For a matched pair(G1 k ,G2 k )with eigenvalues λ 1,k and λ 2,k, the shape loss is, \begin {aligned} \loss {shape}^{(k)} &= \mathcal {H}\!\left (\left \|\log \boldsymbol {\lambda }^{1,k}… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison on ScanNet++ [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Cross-dataset generalization: ScanNet++→DTU. Compared to other generalizable 3DGS methods, MAC-Splat better preserves geometry and edges and produces fewer artifacts under the controlled lighting conditions of DTU. MAC-Splat improves the average PSNR across the four subsets by more than 4.5 dB and reduces the average LPIPS by about 44%. Robustness to wide baselines. A key trend in [PITH_FULL_IMAGE:figures… view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative ablation on ScanNet++. Pure 2D photometric supervision (w/o MAC Loss) yields severe geometric distortions and floaters. While the MAC loss alone (w/o DINOv3 ) corrects global 3D geometry, our full model uses semantically enriched correspondences to further refine local details and cleanly preserve thin struc￾tures. Splat reaches 19.06 dB, yielding a +6.02 dB gap in regions that are jointly ob￾s… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

69 extracted references · 17 linked inside Pith

  1. [1]

    In: ICCV

    Barron, J.T., Mildenhall, B., Tancik, M., Hedman, P., Martin-Brualla, R., Srini- vasan, P.P.: Mip-nerf: A multiscale representation for anti-aliasing neural radiance fields. In: ICCV. pp. 5855–5864 (2021)

    2021

  2. [2]

    Charatan,D.,Li,S.L.,Tagliasacchi,A.,Sitzmann,V.:pixelsplat:3dgaussiansplats fromimagepairsforscalablegeneralizable3dreconstruction.In:CVPR.pp.19457– 19467 (2024)

    2024

  3. [3]

    arXiv preprint arXiv:2401.03890 (2025)

    Chen, G., Wang, W.: A survey on 3d gaussian splatting. arXiv preprint arXiv:2401.03890 (2025)

    Pith/arXiv arXiv 2025

  4. [4]

    In: Proceedings of the 33rd ACM International Conference on Multimedia

    Chen, K., Dai, B., Qin, M., Zhang, D., Li, P., Zou, Y., Wang, H.: Slgaussian: Fast language gaussian splatting in sparse views. In: Proceedings of the 33rd ACM International Conference on Multimedia. pp. 3047–3056 (2025)

    2025

  5. [5]

    In: ECCV

    Chen, Y., Xu, H., Zheng, C., Zhuang, B., Pollefeys, M., Geiger, A., Cham, T.J., Cai, J.: Mvsplat: Efficient 3d gaussian splatting from sparse multi-view images. In: ECCV. pp. 370–386. Springer (2024) 16 J. Yang et al

    2024

  6. [6]

    In: CVPR

    Chung, J., Oh, J., Lee, K.M.: Depth-regularized optimization for 3d gaussian splat- ting in few-shot images. In: CVPR. pp. 811–820 (2024)

    2024

  7. [7]

    IEEE Access12, 96797–96820 (2024)

    Dalal, A., Hagen, D., Robbersmyr, K.G., Knausgård, K.M.: Gaussian splatting: 3d reconstruction and novel view synthesis: A review. IEEE Access12, 96797–96820 (2024)

    2024

  8. [8]

    arXiv preprint arXiv:2408.04831 (2024)

    Du, B., Meng, L., Hu, W.: Auggs: Self-augmented gaussians with structural masks for sparse-view 3d reconstruction. arXiv preprint arXiv:2408.04831 (2024)

    Pith/arXiv arXiv 2024

  9. [9]

    arXiv preprint arXiv:2403.20309 (2024)

    Fan, Z., Cong, W., Wen, K., Wang, K., Zhang, J., Ding, X., Xu, D., Ivanovic, B., Pavone, M., Pavlakos, G., et al.: Instantsplat: Sparse-view gaussian splatting in seconds. arXiv preprint arXiv:2403.20309 (2024)

    Pith/arXiv arXiv 2024

  10. [10]

    In: CVPR

    Guédon, A., Lepetit, V.: Sugar: Surface-aligned gaussian splatting for efficient 3d mesh reconstruction and high-quality mesh rendering. In: CVPR. pp. 5354–5363 (2024)

    2024

  11. [11]

    NeurIPS37, 68595–68621 (2024)

    Han, L., Zhou, J., Liu, Y.S., Han, Z.: Binocular-guided 3d gaussian splatting with view consistency for sparse view synthesis. NeurIPS37, 68595–68621 (2024)

    2024

  12. [12]

    arXiv preprint arXiv:2503.08093 (2025)

    Hou, C., Yeo, Q.X., Guo, M., Su, Y., Li, Y., Lee, G.H.: Mvgsr: Multi-view consistency gaussian splatting for robust surface reconstruction. arXiv preprint arXiv:2503.08093 (2025)

    Pith/arXiv arXiv 2025

  13. [13]

    In: Proceedings of the 33rd ACM International Conference on Multimedia

    Huang, C., Zhang, Q., Zhang, Q., Li, N., Gong, Y., Wang, X., Feng, W.: Trigs: Tri- consistency 3d gaussian splatting from sparse and unposed views. In: Proceedings of the 33rd ACM International Conference on Multimedia. pp. 699–708 (2025)

    2025

  14. [14]

    In: AAAI

    Huang, H., Wu, Y., Deng, C., Gao, G., Gu, M., Liu, Y.S.: Fatesgs: Fast and accu- rate sparse-view surface reconstruction using gaussian splatting with depth-feature consistency. In: AAAI. pp. 3644–3652 (2025)

    2025

  15. [15]

    In: Proceedings of the IEEE/CVF International Con- ference on Computer Vision

    Huang, R., Mikolajczyk, K.: No pose at all: Self-supervised pose-free 3d gaussian splatting from sparse views. In: Proceedings of the IEEE/CVF International Con- ference on Computer Vision. pp. 27947–27957 (2025)

    2025

  16. [16]

    arXiv preprint arXiv:2505.02175 (2025)

    Jena, S., Vutukur, S.R., Boukhayma, A.: Sparsplat: Fast multi-view reconstruction with generalizable 2d gaussian splatting. arXiv preprint arXiv:2505.02175 (2025)

    Pith/arXiv arXiv 2025

  17. [17]

    In: CVPR

    Jensen, R., Dahl, A., Vogiatzis, G., Tola, E., Aanæs, H.: Large scale multi-view stereopsis evaluation. In: CVPR. pp. 406–413 (2014)

    2014

  18. [18]

    ACM Transactions on Graphics (TOG)44(6), 1–16 (2025)

    Jiang, L., Mao, Y., Xu, L., Lu, T., Ren, K., Jin, Y., Xu, X., Yu, M., Pang, J., Zhao, F., et al.: Anysplat: Feed-forward 3d gaussian splatting from unconstrained views. ACM Transactions on Graphics (TOG)44(6), 1–16 (2025)

    2025

  19. [19]

    In: CVPR

    Keetha, N., Karhade, J., Jatavallabhula, K.M., Yang, G., Scherer, S., Ramanan, D., Luiten, J.: Splatam: Splat track & map 3d gaussians for dense rgb-d slam. In: CVPR. pp. 21357–21366 (2024)

    2024

  20. [20]

    ACM Transactions on Graphics42(4), 139:1–139:14 (2023)

    Kerbl, B., Kopanas, G., Leimkühler, T., Drettakis, G.: 3D Gaussian Splatting for Real-Time Radiance Field Rendering. ACM Transactions on Graphics42(4), 139:1–139:14 (2023)

    2023

  21. [21]

    In: CVPR

    Kong, H., Yang, X., Wang, X.: Generative sparse-view gaussian splatting. In: CVPR. pp. 26745–26755 (2025)

    2025

  22. [22]

    In: ECCV

    Kumar, R., Vats, V.: Few-shot novel view synthesis using depth aware 3d gaussian splatting. In: ECCV. pp. 1–13. Springer (2024)

    2024

  23. [23]

    In: ECCV

    Leroy, V., Cabon, Y., Revaud, J.: Grounding image matching in 3d with mast3r. In: ECCV. pp. 71–91. Springer (2024)

    2024

  24. [24]

    In: European conference on computer vision

    Li, Y., Lyu, C., Di, Y., Zhai, G., Lee, G.H., Tombari, F.: Geogaussian: Geometry- aware gaussian splatting for scene rendering. In: European conference on computer vision. pp. 441–457. Springer (2024) MAC-Splat 17

    2024

  25. [25]

    arXiv preprint arXiv:2503.18052 (2025)

    Li, Y., Ma, Q., Yang, R., Li, H., Ma, M., Ren, B., Popovic, N., Sebe, N., Konukoglu, E., Gevers, T., Gool, L.V., Oswald, M.R., Paudel, D.P.: Scenesplat: Gaussian splatting-based scene understanding with vision-language pretraining. arXiv preprint arXiv:2503.18052 (2025)

    Pith/arXiv arXiv 2025

  26. [26]

    In: Proceedings of the IEEE/CVF In- ternational Conference on Computer Vision (ICCV) Workshops

    Liu, C., Ouyang, C., Chen, Y., Quilodrán-Casas, C., Ma, L., Fu, J., Guo, Y., Shah, A., Bai, W., Arcucci, R.: T3d: Advancing 3d medical vision-language pre-training by learning multi-view visual consistency. In: Proceedings of the IEEE/CVF In- ternational Conference on Computer Vision (ICCV) Workshops. pp. 6763–6773 (October 2025)

    2025

  27. [27]

    In: ECCV

    Liu, T., Wang, G., Hu, S., Shen, L., Ye, X., Zang, Y., Cao, Z., Li, W., Liu, Z.: Mvsgaussian: Fast generalizable gaussian splatting reconstruction from multi-view stereo. In: ECCV. pp. 37–53. Springer (2024)

    2024

  28. [28]

    NeurIPS37, 133305–133327 (2024)

    Liu, X., Zhou, C., Huang, S.: 3dgs-enhancer: Enhancing unbounded 3d gaussian splatting with view-consistent 2d diffusion priors. NeurIPS37, 133305–133327 (2024)

    2024

  29. [29]

    Image and Vision Computing151, 105304 (2024)

    Luo,J.,Huang,T.,Wang,W.,Feng,W.:Areviewofrecentadvancesin3dgaussian splatting for optimization and reconstruction. Image and Vision Computing151, 105304 (2024)

    2024

  30. [30]

    ACM Transac- tions on Graphics (TOG)43(6), 1–12 (2024)

    Lyu, X., Sun, Y.T., Huang, Y.H., Wu, X., Yang, Z., Chen, Y., Pang, J., Qi, X.: 3dgsr: Implicit surface reconstruction with 3d gaussian splatting. ACM Transac- tions on Graphics (TOG)43(6), 1–12 (2024)

    2024

  31. [31]

    In: CVPR

    Matsuki, H., Murai, R., Kelly, P.H., Davison, A.J.: Gaussian splatting slam. In: CVPR. pp. 18039–18048 (2024)

    2024

  32. [32]

    Commu- nications of the ACM65(1), 99–106 (2021)

    Mildenhall, B., Srinivasan, P.P., Tancik, M., Barron, J.T., Ramamoorthi, R., Ng, R.: Nerf: Representing scenes as neural radiance fields for view synthesis. Commu- nications of the ACM65(1), 99–106 (2021)

    2021

  33. [33]

    Neurocomputing438, 14–33 (2021)

    Ming,Y.,Meng,X.,Fan,C.,Yu,H.:Deeplearningformonoculardepthestimation: A review. Neurocomputing438, 14–33 (2021)

    2021

  34. [34]

    In: CVPR

    Park, H., Ryu, G., Kim, W.: Dropgaussian: Structural regularization for sparse- view gaussian splatting. In: CVPR. pp. 21600–21609 (2025)

    2025

  35. [35]

    NeurIPS 37, 97328–97352 (2024)

    Peng, R., Xu, W., Tang, L., Liao, L., Jiao, J., Wang, R.: Structure consistent gaussian splatting with matching prior for few-shot novel view synthesis. NeurIPS 37, 97328–97352 (2024)

    2024

  36. [36]

    In: CVPR

    Qin, M., Li, W., Zhou, J., Wang, H., Pfister, H.: Langsplat: 3d language gaussian splatting. In: CVPR. pp. 20051–20060 (2024)

    2024

  37. [37]

    In: CVPR

    Schönberger, J.L., Frahm, J.M.: Structure-from-motion revisited. In: CVPR. pp. 4104–4113 (2016)

    2016

  38. [38]

    Computers & Graphics p

    Shen, J., Feng, T., Dong, H., Ji, J., Wang, X., Shao, T.: Surags: Toward efficient few-shot novel view synthesis via surface-aware gaussian splatting. Computers & Graphics p. 104349 (2025)

    2025

  39. [39]

    arXiv preprint arXiv:2508.10104 (2025)

    Siméoni, O., Vo, H.V., Seitzer, M., Baldassarre, F., Oquab, M., Jose, C., Khali- dov, V., Szafraniec, M., Yi, S., Ramamonjisoa, M., et al.: Dinov3. arXiv preprint arXiv:2508.10104 (2025)

    Pith/arXiv arXiv 2025

  40. [40]

    arXiv preprint arXiv:2408.13912 (2024)

    Smart, B., Zheng, C., Laina, I., Prisacariu, V.A.: Splatt3r: Zero-shot gaussian splatting from uncalibrated image pairs. arXiv preprint arXiv:2408.13912 (2024)

    Pith/arXiv arXiv 2024

  41. [41]

    In: ICCV

    Sucar, E., Liu, S., Ortiz, J., Davison, A.J.: imap: Implicit mapping and positioning in real-time. In: ICCV. pp. 6229–6238 (2021)

    2021

  42. [42]

    arXiv preprint arXiv:2505.20729 (2025) 18 J

    Sun, X., Chen, R., Gong, M., Xu, D., Liu, T.: Intern-gs: Vision model guided sparse-view 3d gaussian splatting. arXiv preprint arXiv:2505.20729 (2025) 18 J. Yang et al

    Pith/arXiv arXiv 2025

  43. [43]

    In: CVPR

    Tancik, M., Casser, V., Yan, X., Pradhan, S., Mildenhall, B., Srinivasan, P.P., Bar- ron, J.T., Kretzschmar, H.: Block-nerf: Scalable large scene neural view synthesis. In: CVPR. pp. 8248–8258 (2022)

    2022

  44. [44]

    arXiv preprint arXiv:2410.06245 (2024)

    Tang, S., Ye, W., Ye, P., Lin, W., Zhou, Y., Chen, T., Ouyang, W.: Hisplat: Hier- archical 3d gaussian splatting for generalizable sparse-view reconstruction. arXiv preprint arXiv:2410.06245 (2024)

    Pith/arXiv arXiv 2024

  45. [45]

    In: 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)

    Turkulainen, M., Ren, X., Melekhov, I., Seiskari, O., Rahtu, E., Kannala, J.: Dn- splatter: Depth and normal priors for gaussian splatting and meshing. In: 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV). pp. 2421–2431. IEEE (2025)

    2025

  46. [46]

    In: ECCV

    Ververas, E., Potamias, R.A., Song, J., Deng, J., Zafeiriou, S.: Sags: Structure- aware 3d gaussian splatting. In: ECCV. pp. 221–238. Springer (2024)

    2024

  47. [47]

    In: CVPR

    Wang, S., Leroy, V., Cabon, Y., Chidlovskii, B., Revaud, J.: Dust3r: Geometric 3d vision made easy. In: CVPR. pp. 20697–20709 (2024)

    2024

  48. [48]

    NeurIPS37, 107326–107349 (2024)

    Wang, Y., Huang, T., Chen, H., Lee, G.H.: Freesplat: Generalizable 3d gaussian splatting towards free view synthesis of indoor scenes. NeurIPS37, 107326–107349 (2024)

    2024

  49. [49]

    arXiv preprint arXiv:2410.04738 (2024)

    Wang, Z., Li, D., Wu, Y., He, T., Bian, J., Jiang, R.: Diffusion models in 3d vision: A survey. arXiv preprint arXiv:2410.04738 (2024)

    Pith/arXiv arXiv 2024

  50. [50]

    arXiv preprint arXiv:2504.00639 (2025)

    Wu,J.,Li,H.,Jiang,X.,Yao,Y.,Zhang,L.:Coca-splat:Collaborativeoptimization for camera parameters and 3d gaussians. arXiv preprint arXiv:2504.00639 (2025)

    Pith/arXiv arXiv 2025

  51. [51]

    Computational Visual Media10(4), 613–642 (2024)

    Wu, T., Yuan, Y.J., Zhang, L.X., Yang, J., Cao, Y.P., Yan, L.Q., Gao, L.: Recent advances in 3d gaussian splatting. Computational Visual Media10(4), 613–642 (2024)

    2024

  52. [52]

    arXiv preprint arXiv:2312.00206 (2023)

    Xiong, H., Muttukuru, S., Xiao, H., Upadhyay, R., Chari, P., Zhao, Y., Kadambi, A.: SparseGS: Sparse View Synthesis using 3D Gaussian Splatting. arXiv preprint arXiv:2312.00206 (2023)

    Pith/arXiv arXiv 2023

  53. [53]

    In: ICCV

    Xu, J., Gao, S., Shan, Y.: Freesplatter: Pose-free gaussian splatting for sparse-view 3d reconstruction. In: ICCV. pp. 25442–25452 (2025)

    2025

  54. [54]

    In: ECCV

    Xu, Y., Shi, Z., Yifan, W., Chen, H., Yang, C., Peng, S., Shen, Y., Wetzstein, G.: Grm: Large gaussian reconstruction model for efficient 3d reconstruction and generation. In: ECCV. pp. 1–20. Springer (2024)

    2024

  55. [55]

    In: CVPR

    Yan, C., Qu, D., Xu, D., Zhao, B., Wang, Z., Wang, D., Li, X.: Gs-slam: Dense visual slam with 3d gaussian splatting. In: CVPR. pp. 19595–19604 (2024)

    2024

  56. [56]

    arXiv preprint arXiv:2402.10259 (2024)

    Yang, C., Li, S., Fang, J., Liang, R., Xie, L., Zhang, X., Shen, W., Tian, Q.: Gaus- sianobject: High-quality 3d object reconstruction from four views with gaussian splatting. arXiv preprint arXiv:2402.10259 (2024)

    Pith/arXiv arXiv 2024

  57. [57]

    In: ECCV

    Yao, Y., Luo, Z., Li, S., Fang, T., Quan, L.: Mvsnet: Depth inference for unstruc- tured multi-view stereo. In: ECCV. pp. 767–783 (2018)

    2018

  58. [58]

    arXiv preprint arXiv:2410.24207 (2024)

    Ye, B., Liu, S., Xu, H., Li, X., Pollefeys, M., Yang, M.H., Peng, S.: No pose, no problem: Surprisingly simple 3d gaussian splats from sparse unposed images. arXiv preprint arXiv:2410.24207 (2024)

    Pith/arXiv arXiv 2024

  59. [59]

    In: ICCV

    Yeshwanth, C., Liu, Y.C., Nießner, M., Dai, A.: Scannet++: A high-fidelity dataset of 3d indoor scenes. In: ICCV. pp. 12–22 (2023)

    2023

  60. [60]

    NeurIPS37, 127204–127225 (2024)

    Yin, R., Yugay, V., Li, Y., Karaoglu, S., Gevers, T.: Fewviewgs: Gaussian splatting with few view matching and multi-stage training. NeurIPS37, 127204–127225 (2024)

    2024

  61. [61]

    arXiv preprint arXiv:2507.16406 (2025)

    Younis, T., Cheng, Z.: Sparse-view 3d reconstruction: Recent advances and open challenges. arXiv preprint arXiv:2507.16406 (2025)

    Pith/arXiv arXiv 2025

  62. [62]

    arXiv preprint arXiv:2409.03456 (2024) MAC-Splat 19

    Yu, H., Long, X., Tan, P.: Lm-gaussian: Boost sparse-view 3d gaussian splatting with large model priors. arXiv preprint arXiv:2409.03456 (2024) MAC-Splat 19

    Pith/arXiv arXiv 2024

  63. [63]

    In: CVPR

    Yu, Z., Chen, A., Huang, B., Sattler, T., Geiger, A.: Mip-splatting: Alias-free 3d gaussian splatting. In: CVPR. pp. 19447–19456 (2024)

    2024

  64. [64]

    In: European conference on computer vision

    Zhang, J., Li, J., Yu, X., Huang, L., Gu, L., Zheng, J., Bai, X.: Cor-gs: sparse-view 3d gaussian splatting via co-regularization. In: European conference on computer vision. pp. 335–352. Springer (2024)

    2024

  65. [65]

    In: CVPR

    Zhang, R., Isola, P., Efros, A.A., Shechtman, E., Wang, O.: The unreasonable effectiveness of deep features as a perceptual metric. In: CVPR. pp. 586–595 (2018)

    2018

  66. [66]

    In: ICCV

    Zhao, C., Wang, X., Zhang, T., Javed, S., Salzmann, M.: Self-ensembling gaussian splatting for few-shot novel view synthesis. In: ICCV. pp. 4940–4950 (2025)

    2025

  67. [67]

    In: CVPR

    Zheng, S., Zhou, B., Shao, R., Liu, B., Zhang, S., Nie, L., Liu, Y.: Gps-gaussian: Generalizable pixel-wise 3d gaussian splatting for real-time human novel view syn- thesis. In: CVPR. pp. 19680–19690 (2024)

    2024

  68. [68]

    In: ECCV

    Zhu, Z., Fan, Z., Jiang, Y., Wang, Z.: Fsgs: Real-time few-shot view synthesis using gaussian splatting. In: ECCV. pp. 145–163. Springer (2024)

    2024

  69. [69]

    In: CVPR

    Zou, Z.X., Yu, Z., Guo, Y.C., Li, Y., Liang, D., Cao, Y.P., Zhang, S.H.: Triplane meets gaussian splatting: Fast and generalizable single-view 3d reconstruction with transformers. In: CVPR. pp. 10324–10335 (2024)

    2024